← Back to all companies
M

Mistral AI

French AI startup releasing open-weight models; a leading European frontier AI company.

Safety Documents

⚠️
Responsible Scaling Policy
Mistral AI Safety Policy / Acceptable Use Policy. No comprehensive Responsible Scaling Policy equivalent found. Mistral has usage policies and a general safety approach. No FSF-equivalent framework published.🔗
⚠️
Model Card / System Card
Mistral model technical reports / Hugging Face cards. Technical documentation on Hugging Face for open-weight models. No formal system cards with safety evaluation sections comparable to frontier labs found.🔗
❌
Safety Benchmark Results
No published comprehensive safety benchmark results found in frontier lab standard format. Enkrypt AI study found Pixtral models significantly more vulnerable than OpenAI models.
✅
Acceptable Use Policy
Mistral has Terms of Service and AUP. Coverage of prohibited uses less detailed than major US labs.🔗

Testing & Evaluation

✅
Third-Party Red Teaming
Enkrypt AI (independent — published May 2025). Enkrypt AI independently red-teamed Pixtral-Large (25.02) and Pixtral-12b. Found 60x more likely to produce CSEM than OpenAI and 18-40x more likely to produce CBRN information. Not commissioned by Mistral.🔗
⚠️
CBRN Risk Evaluation
No published CBRN-specific evaluation by Mistral found. Enkrypt AI found Pixtral models 18-40x more likely to produce CBRN content than other models.
⚠️
Pre-Deployment Safety Evaluation
No published pre-deployment safety evaluation in standard format found.
❌
External Safety Evaluations
No evidence of participation in METR, ARC, UK AISI or other formal external evaluation programs.

Governance

❌
Independent Safety Board
No publicly documented independent safety board found.
✅
Seoul AI Safety Commitment
Mistral AI is explicitly named as one of the 16 Seoul Commitment signatories per AP News and UK government source.🔗
Last updated: 2024-05-21
⚠️
Government Safety Report
EU AI Act compliance (GPAI Code of Practice). Mistral has signed the EU GPAI Code of Practice. As a French company, subject to EU AI Act. CEO Arthur Mensch has engaged with EU regulators.🔗
⚠️
Third-Party Audits
No evidence of AI safety-specific third-party audits published.
⚠️
Whistleblower Policy
No publicly documented whistleblower or safety concern process found.

Policy Positions

Military Use:unknown
No explicit comprehensive military use policy found in public documentation.🔗
Surveillance Use:unknown
No explicit surveillance policy found in public documentation.🔗
Open-Source Models:Yes
Mistral 7B (Apache 2.0), Mixtral 8x7B (Apache 2.0), Mixtral 8x22B (Apache 2.0), Mistral NeMo 12B (Apache 2.0), Mistral Small 3 (Apache 2.0). and 2 more. Many Mistral models are open-weight under Apache 2.0. Closed-source commercial models also exist (Mistral Large, Pixtral Large API-only).🔗
Children/Minors Policy:Unknown
No prominently documented children/minors policy found. Enkrypt AI study found Pixtral models 60x more likely to produce CSEM than OpenAI models.

Incident History

Enkrypt AI: Mistral Pixtral Models 60x More Likely to Generate CSEM Than OpenAI

2025-05-08

US-based AI security company Enkrypt AI published a report finding that Mistral's Pixtral-Large (25.02) and Pixtral-12b models were 60 times more prone to generate child sexual exploitation material (CSEM) than OpenAI models, and 18-40 times more likely to produce dangerous CBRN information under adversarial prompting.

🔗 Source🔗 Company Response

Mistral Models Found to Reproduce Copyrighted Text

2024-03

Research in March 2024 found Mistral models reproduced verbatim copyrighted text in 44%, 22%, 10%, and 8% of responses depending on model. This raised intellectual property and training data safety concerns.

🔗 Source🔗 Company Response

Mistral 7B Released with No Safety Fine-Tuning

2023-09

Mistral's first public model, Mistral 7B, was released without standard safety fine-tuning, allowing it to respond to harmful requests more readily than safety-tuned alternatives. This positioned it as an 'uncensored' alternative but drew criticism from safety researchers.

🔗 Source🔗 Company Response

Timeline

2025-12-02
Mistral 3 family released as open-weight frontier models🔗
2025-05-08
Enkrypt AI publishes report on CSEM and CBRN vulnerabilities in Mistral Pixtral models🔗
2024-05-21
Mistral AI signs Seoul Frontier AI Safety Commitments🔗
2024-01-11
Mixtral 8x7B released under Apache 2.0🔗
2023-09-27
Mistral 7B released under Apache 2.0 — first open-weight model from Mistral🔗